摘要
机采茶鲜叶的叶和芽混合,利用茶叶图像纹理特征对茶鲜叶进行分类,分类之后再加工有助于提高茶叶的质量和市场价值。支持向量机SVM是一种专门针对小样本、非线性、高维特征的经典分类算法,但对于茶叶这类自然图片在分界面附近的测试点容易出错。KNN是一种简单而经典的分类算法,核心在于向量间距离的计算,论文提出欧式距离和余弦相似度结合的方式作为KNN新的距离计算公式。改进的KNN与SVM结合起来,形成SVM-KNN算法应用于茶叶图像的纹理特征分类的研究中,并分析SVM-KNN的时间复杂度。对比实验表明,SVMKNN算法对茶叶图像纹理分类正确率有很大程度地提高,最高可达90%以上。
The mechanical plucking tea are mixture of leaves and buds,thus if tea leaves can be classified by the texture feature of tea images before processing,it will contribute to improving the quality and market value of tea leaves.Support Vector Machine is a specific assort algorithm for small samples,nonlinear and high dimensional,but it easily makes mistakes near the hyperplane for nature image such as tea images. K-Nearest-Neighbor is a classic classification algorithm whose key point is to calculate the absolute space distance between different vectors. This paper proposed a new distance formula that linearly combined Euclidean with Cosine Similarity,and then,used the improved KNN and SVM to form a new algorithm called SVM-KNN which would be applied to the research of texture classification of tea images. Besides,it analyzed the time complexity of SVM. The comparison experiments showed that SVM-KNN obviously increased the accuracy of tea images assortment up to 90% and above.
出处
《中国茶叶加工》
2016年第6期5-9,共5页
China Tea Processing
基金
国家自然科学基金(31470028)
湖南省战略性新兴产业科技攻关项目(2014GK1020)
